TCPMPL Interface

TCPMPL has been implemented in C++. It gives the features of TCPMPL as a class interface to the user application which uses it. The following is the class interface:

 
  class TCPMPL {

  		TCPMPL(void);

  		TCPMPL();

  		void		initTCPMPL(int &argc, char ** &argv);

  		int		getId(void) const;

  		int		getTCPMPLSize()() const;

  		bool		sendData(char *ptr, int nbytes, int id);

  		bool		sendDataVector(const struct iovec *vector, int count,
  int id);

  		bool		recvDataFrom(char *ptr, int nbytes, int id);

  		bool		recvData(char *ptr, int nbytes, int *id);

  		bool		probeForData(int *source, int *length, long seconds,
  long microseconds);

  		bool		waitForData();

  }

initTCPMPL
This is the method that user applications need to call before any user code that can be executed in parallel. This method starts the processes on the remote machines specified. It also establishes the communication channels between all its peers.

getId
This method is used to get the identity of the parallel process. It returns a value 0 for the master and 1 to N-1 for the slaves.

getTCPMPLSize
This method is used by user applications to get the total number of processes running in the group. It returns N, where N is the total number of processes in the group.
sendData
This method is used by user application to send information to its peers. This method takes as input the information to be sent as a sequence of bytes and the length of the buffer to be sent together with the identity of the destination process in the group.
sendDataVector
This method is used to send multiple buffers of data as a single message to the destination process. Sometimes user applications have information stored in multiple buffers. All the data in multiple buffers need to be sent to the destination as a single message. This reduces the copying since the user application does not need to copy all the data into a single contiguous buffer to send the data to the destination process.
recvDataFrom
This method is used by user applications to receive data from a particular process identified by the identity number. It also has to specify the size of the message to be received. If the data that is waiting to be received is larger than the size specified by the user, the routine returns error.
recvData
This method is used by user application to receive data from any source process. This method requires the user application to specify the size of the message that has to be received. This method returns the identity of the process from which the data has been received back to the user.
probeForData
This method is used by user applications to probe for any data that is available from any source process. Asynchronous user applications do not know when the data is arriving to a particular process. Therefore, this method is used to check whether there is any message that has arrived from any source process. It returns the message size and the identity of the process from which the message is available back to the user application. It also asks the user application to specify the time in seconds and microseconds for which probeForData has to wait before returning control back to the user about the status of message available.
waitForData
This method is used by user applications to wait for data to arrive from any source process. This method is a blocking call. It waits infinitely for a message to arrive. User applications are advised to use this call only if the application knows for sure that a message is going to arrive to this process.